Occasionally meta-analyses use ‘vote counting’ to compare the number of positive studies with the number of negative studies. Vote counting is limited to answering the simple question “is there any evidence of an effect?” Two problems can occur with vote counting, which suggest that it should be avoided whenever possible. Firstly, problems occur if subjective decisions or statistical significance are used to define ‘positive’ and ‘negative’ studies (Cooper 1980, Antman 1992). To undertake vote counting properly the number of studies showing harm should be compared with the number showing benefit, regardless of the statistical significance or size of their results. A sign test can be used to assess the significance of evidence for the existence of an effect in either direction (if there is no effect the studies will be distributed evenly around the null hypothesis of no difference). Secondly, vote counting takes no account of the differential weights given to each study. Vote counting might be considered as a last resort in situations when standard meta-analytical methods cannot be applied (such as when there is no consistent outcome measure).